AI agent vs. chatbot vs. workflow: which one do you need?
A chatbot answers. A workflow repeats. An agent decides. Here is how to choose.
Every AI product now says it has agents. Most business owners do not need to know how agents are built. They need to know which of three things fits the task in front of them.
The three in one sentence each
- A chatbot answers the message you send, then waits for the next one.
- A workflow runs the same fixed steps each time, with AI doing one or more of the steps.
- An AI agent gets a goal, decides the steps on its own, uses tools and checks its progress.
Side by side
| Chatbot | Workflow | AI agent | |
|---|---|---|---|
| Who decides the next step | You | The fixed process | The AI |
| Best for | Thinking, drafting, one-off questions | Tasks you repeat the same way | Tasks where each case differs |
| Uses tools | Sometimes | Yes, the ones you set | Yes, and it chooses which |
| Predictable | High, you see each step | High | Lower |
| Your effort per task | High | Low after setup | Low, plus review |
| Main risk | It takes your time | It breaks on unusual cases | It repeats an early error |
One task, three ways
Take a simple task. A new lead fills in your contact form, and you want to reply well.
With a chatbot, you paste the form into ChatGPT or Claude and ask for a reply. You read it, edit it and send it. Good result, and ten minutes of your time for each lead.
With a workflow, each new form starts the same steps. The AI drafts a reply from your template, and the draft lands in your inbox for approval. You spend one minute per lead. An unusual request still gets the standard reply.
With an agent, the AI reads the form, looks up the company, checks your calendar, picks the right offer and drafts a reply that fits this lead. It handles the unusual request. It can also get the company wrong, so you review before it sends.
A rule for choosing
- If you do the task less than one time per week, use a chatbot.
- If you do it often and the steps are the same, build an agentic workflow with fixed steps.
- If you do it often and each case needs judgment, use an agent, with a human in the loop before anything is sent.
Start one level lower than you think you need. A workflow you trust beats an agent you have to watch.
What to set before you delegate
The more an AI system does alone, the more its limits matter. Decide what it may read, what it may change and what needs your approval. These limits are called guardrails. Be careful with agents that read email or web pages, because of prompt injection.
For agents that improve their own work, see 20 loop prompts for a solo business. For the numbers on how often agents complete real tasks, see the AI reliability statistics. To move from a chatbot to agents in order, read the four steps past the chat level of AI.